CN110994673B - Prediction method for micro-grid self-adaptive anti-islanding disturbance load impedance value - Google Patents
Prediction method for micro-grid self-adaptive anti-islanding disturbance load impedance value Download PDFInfo
- Publication number
- CN110994673B CN110994673B CN201911170555.XA CN201911170555A CN110994673B CN 110994673 B CN110994673 B CN 110994673B CN 201911170555 A CN201911170555 A CN 201911170555A CN 110994673 B CN110994673 B CN 110994673B
- Authority
- CN
- China
- Prior art keywords
- load impedance
- disturbance load
- islanding
- impedance value
- fault
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/38—Arrangements for parallely feeding a single network by two or more generators, converters or transformers
- H02J3/381—Dispersed generators
Landscapes
- Engineering & Computer Science (AREA)
- Power Engineering (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention relates to a method for predicting a self-adaptive anti-islanding disturbance load impedance value of a micro-grid, which comprises the following steps: 1) judging whether to predict a disturbance load impedance value according to the states of the anti-islanding device in the microgrid under different application scenes; 2) when the micro-grid operates normally before the anti-islanding device is put into use, the disturbance load impedance value is predicted through the LSTM; 3) when the micro-grid fails before the anti-islanding device is put into use, protective action information is collected, and after a micro-grid fault information extraction model is established based on a Petri network, a disturbance load impedance value is predicted through an LSTM. Compared with the prior art, the method has the advantages of considering the impedance characteristics under multiple scenes, enhancing the flexibility and reliability of the operation of the micro-grid and the like.
Description
Technical Field
The invention relates to the field of micro-grid operation and maintenance, in particular to a micro-grid self-adaptive anti-islanding disturbance load impedance value prediction method.
Background
At present, researches on anti-islanding devices are developed from aspects of zero blind area detection, blind area prediction, power grid self-healing capacity, self-adaption removal and the like, few researches are conducted on the problems of determination of anti-islanding equivalent disturbance load impedance values, the existing fixed impedance values cannot be adjusted in real time according to supply and demand changes in a microgrid, and the flexibility and reliability of power grid operation are reduced. Therefore, the method for researching the adaptive anti-islanding disturbance load and the variable resistance value has important significance.
Under the normal operation condition, the disturbance load impedance value is greatly influenced by the time sequence, and the output of the distributed power supply in the microgrid is related to the external weather environment change and the power supply output state at the last moment, so that the time sequence characteristics are fully considered in the impedance value selection process, and the controllability of the memory information on the time sequence is realized. Once a fault occurs, the total impedance value of the microgrid system is discretely and parallelly changed under the influence of the impedance value of a fault line, and the input disturbance load impedance value dynamically changes to realize real-time response to the supply and demand relationship in the microgrid. At present, a study method for disturbance load impedance is mainly considered from the aspects of adaptive impedance angle, impedance feature extraction based on wavelet basis selection and relative wave impedance prediction precision, and feature analysis under different scenes of disturbance load impedance values and design of an adaptive prediction method need further study in a prediction process.
With the gradual increase of permeability of distributed power supplies, the influence on limiting the maximum capacity of new energy which can be accessed, guaranteeing the safety of maintainers, ensuring the normal work of a power grid automatic device, constructing a clean low-carbon energy supply system and the like is increasingly prominent, and because impedance characteristic analysis under different scenes is lacked in the current input process, the reliability and flexibility of system operation are reduced, the method for predicting the self-adaptive anti-island disturbance load impedance value of the micro-grid by combining the LSTM neural network and the Petri network has important significance.
Disclosure of Invention
The invention aims to overcome the defects in the prior art and provide a method for predicting a microgrid adaptive anti-islanding disturbance load impedance value.
The purpose of the invention can be realized by the following technical scheme:
a prediction method for a micro-grid adaptive anti-islanding disturbance load impedance value comprises the following steps:
1) judging whether to predict a disturbance load impedance value according to the states of the anti-islanding device in the microgrid under different application scenes;
2) when the micro-grid operates normally before the anti-islanding device is put into use, the disturbance load impedance value is predicted through the LSTM;
3) when the micro-grid fails before the anti-islanding device is put into use, protective action information is collected, and after a micro-grid fault information extraction model is established based on a Petri network, a disturbance load impedance value is predicted through an LSTM.
In the step 1), the application scenario of the anti-islanding device in the microgrid comprises:
(1) in the island operation maintenance scene, the anti-island device is in an input state, all converter switches in the island are disconnected after the input, and the island stops operating;
(2) under the condition of large power grid faults, a scene of partial island operation is allowed, the anti-island device state is removed, and planned island division is carried out;
(3) under the condition of large power grid faults and abnormal scenes appear in the island operation process, the anti-island device is in an input state, all converter switches in the island are disconnected after the input, and the island stops operating.
In the step 1), when the system is in the scenes of (1) and (3), the disturbance load impedance value is predicted.
In the step 2), the predicting of the disturbance load impedance value through the LSTM specifically includes the following steps:
21) constructing an anti-islanding disturbance load impedance prediction model based on an LSTM neural network, and determining the number of model layers and the number of neuron nodes in each layer;
22) training an anti-islanding disturbance load impedance prediction model, and determining optimal network parameters;
23) and predicting the disturbance load impedance value according to the trained anti-islanding disturbance load impedance prediction model.
In the step 21), the anti-islanding disturbance load impedance prediction model takes photovoltaic current, fan current and gas turbine current data as input layers, the number of nodes of the input layers is 3, the disturbance load impedance value is taken as an output layer, and the number of nodes of the output layers is 1.
In the step 21), a training function of each hidden layer in the anti-islanding disturbance load impedance prediction model adopts a self-adaptive learning rate momentum method, an activation function of each layer adopts a logarithmic sigmoid function, and the number of nodes of the hidden layer is 7.
In the step 22), in the process of training the anti-islanding disturbance load impedance prediction model, the loss function is minimized through iteration of a gradient descent method, the target value and the deviation amount are reduced, and finally the optimal network parameters of the model are obtained.
In the step 3), Token is defined as a confidence level of fault information based on the Petri network, the flow of Token represents the transmission of the fault, and a state variable describing the change of the fault mode is used as a transition to construct a microgrid fault information extraction model.
The micro-grid fault information extraction model is specifically expressed as follows:
S＝(P,T；I,O,K,C,M,Ω,α,f,H,U)
wherein, P ═ { P ═ P1,p2,...,pnN > 0 is a finite, non-empty pool node set representing a failure mode in the microgrid, T＝{t1,t2,...,tmIs a finite non-empty transition node set, tmThe method is characterized in that a library node representing a fault state variable is used for reflecting the replacement change of a system fault propagation process, I is a Petri net input matrix and reflects the mapping from a library to a transition, O is a Petri net output matrix and reflects the mapping from the transition to the library, and K is K and { K { (K) } T × P1,k2,...,kn-Token, representing a limited set of fault information, C a set of colors to represent different fault types by a color library, M ═ M (M)1,m2,...,mn)TDistributing n-dimensional vectors, element m, for library identitiesi(i ═ 1, 2.., n) denotes the corresponding library piThe number of tokens and the color of Token, the number of tokens indicates whether a fault occurs, the severity of the fault, and the number of paths causing the fault, and Ω ═ (ω ═ is1,ω2,...,ωn)TN-dimensional vector of library weight, representing the influence of the input library p on the transition rule t, α ═ α1,α2,...,αn)TFor the fault event confidence n-dimensional vector, f ═ f1,f2,...,fn)TIs a set of library incident failure rates, where fiCorresponding to a library p in the fault information extraction modeliAnd represents piSet of failure rates of (H ═ λ)1,λ2,...,λn)TFor the transition rule threshold m-dimensional vector, U ═ diag (μ)1,μ2,...,μm) For transition rule confidence matrix, element mujFor the transition rule tjA threshold value of (d), and μj∈[0,1]。
Compared with the prior art, the invention has the following advantages:
according to the method, the micro-grid adaptive anti-island disturbance load impedance value prediction model is established, the anti-island equivalent impedance application scene and the influence of disturbance load impedance adaptive prediction capability on the micro-grid operation stability can be systematically analyzed, and a clean low-carbon energy supply system is favorably constructed.
Secondly, the invention aims at responding the actual supply and demand relationship in the microgrid in real time, analyzes the impedance characteristics under multiple scenes, predicts the input disturbance load impedance value through an LSTM neural network under the normal condition, firstly establishes a microgrid fault information extraction model by using a Petri network under the condition of outlet line fault, and the LSTM neural network achieves the requirement of 0.001 mean square error value and the minimum mean square error value which can be achieved by the LSTM neural network through 121 times of iterative training in the training process. By continuously adjusting the weight and the threshold in the initial training process, the convergence speed is increased, and the model prediction capability is improved. The self-adaptive anti-islanding prediction model is applied to a micro-grid project, can play a role in protection, and enhances the flexibility and reliability of micro-grid operation.
Drawings
FIG. 1 is a logic diagram of the prediction method of the present invention.
Fig. 2 is a microgrid simulation system in an embodiment.
FIG. 3 shows the results of the LSTM neural network at different hidden nodes.
Fig. 4 is a process of searching for an optimal solution by a gradient descent method.
FIG. 5 is an iteration graph of the LSTM neural network error.
Fig. 6 is a graphical representation of a Petri net fault information extraction-based model.
Detailed Description
The invention is described in detail below with reference to the figures and specific embodiments.
Examples
The invention provides a disturbance load impedance value prediction method based on a Long Short-Term Memory neural network (LSTM), a microgrid fault information extraction model can be established by using a Petri network, and the influence of an anti-islanding equivalent impedance application scene and disturbance load impedance self-adaptive prediction capability on the operation stability of a microgrid is analyzed by a system.
The invention provides a disturbance load impedance input rule aiming at the problem that the disturbance load input under a fixed resistance value cannot be adjusted in real time according to the supply and demand relationship in a microgrid and the operation flexibility of a system is reduced, and multi-scene characteristic analysis and impedance value prediction are carried out on the input process. The application scenes of the anti-islanding device in the microgrid are divided into the following three types:
(1) the island operation is not allowed to be maintained, and the anti-island device state is represented as follows: after the island is put into operation, all converter switches in the island are disconnected, the island stops operating, and the personal safety of maintenance personnel is ensured;
(2) under the condition of large power grid faults, partial isolated island operation is allowed, power failure of users is reduced, and power generation efficiency is improved. The anti-islanding device state is represented as: removing and carrying out planned island division;
(3) under the condition of large power grid faults and when abnormality occurs in the island operation process, the state of the anti-island device is represented as follows: and after the island is put into use, all converter switches in the island are disconnected, the island stops running, and the safety of equipment and personnel is ensured.
Under the (1) and (3) scenes, the anti-islanding device is represented as follows: and (4) putting into a state. If the micro-grid operates normally before the input and no fault occurs, the memory mechanism of the LSTM is utilized, the output value of the last time sequence neuron is referred, and the weight and the offset are continuously updated by adopting a gradient descent method in the reverse propagation process, so that the precision of predicting the input dynamic impedance value is improved. If a fault outgoing line exists in the microgrid before the microgrid is put into operation, a microgrid fault information extraction model is built by utilizing the advantages of strong parallel asynchronism and capability of potential simulation of the Petri network and is used as a data source for predicting the impedance value of the input disturbance load in the scene, then the dynamic impedance value is predicted through the LSTM, and the built prediction model plays an active role in building a clean low-carbon energy supply system. In summary, the logic relationship of the microgrid adaptive anti-islanding disturbance load impedance value prediction method designed by the patent is shown in fig. 1.
A Long Short-Term Memory Neural Network (LSTM) is used as a time Recurrent Neural Network algorithm, and the value of a previous time series neuron is transmitted to a current neuron through a weight matrix on the basis of a standard Recurrent Neural Network (RNN), so that the output of the neuron not only depends on the input of the current time, but also depends on the output of the neuron at the previous time, and the controllability of Memory information on a time series is realized. The memory and forgetting degree of the information before the neuron are respectively controlled by three controllable gates, namely a forgetting Gate (Forget Gate), an Input Gate (Input Gate) and an Output Gate (Output Gate), the retention degree of an Input signal can be judged according to rules, and invalid information is forgotten, so that the neural network has a long-term memory function, and the method has better applicability to disturbance load impedance value prediction on a processing time sequence.
The Petri network provides an analysis method for parallel system research, a correlation matrix is established by combining a Petri network structure and fault information, the initial state of the Petri network is obtained, systematic solving system transition is guaranteed, the state equation management level is improved, and the steady state condition of the Petri network is deduced. And meanwhile, the system can complete concurrent, asynchronous and cyclic work, and realize the comprehensive control strategy of element fault processing, protection action analysis and breaker action fault control. Therefore, a Petri network is adopted to establish a fault information extraction model, and the fault information extraction efficiency and the micro-grid steady-state recovery level are improved.
As shown in figure 2, the test system is characterized in that a part of a simulation micro-grid load component of an IEEE-RBTS BUS main feeder F4 of the test system is selected, three micro-power sources including an equivalent photovoltaic power source, a wind driven generator and a micro gas turbine are added, and a system wiring diagram is shown in figure 2. Wherein L1 and L3 are class 2 loads, L2 and L4 are class 3 loads,
the annual maximum load of the 4 loads is Pmax1＝40.771kW,Pmax2＝33kW，Pmax3＝54kW,Pmax430 kW. The upper limit of the output of a fan (WT) is 20kW, the upper limit of the output of a Photovoltaic (PV) system is 15kW, the lower limit is 5kW, and the upper limit of the output of a micro gas turbine (MT) is 60 kW.
The invention discloses an LSTM neural network-based anti-islanding disturbance load impedance prediction model which mainly comprises a neural network input layer, a hidden layer and an output layer.
The determination method of the number of layers of the neural network model and the number of neuron nodes in each layer is as follows:
(1) determination of the number of input layer nodes:
the photovoltaic power generation system dynamically reflects the real-time running state in the microgrid by collecting photovoltaic current, fan current and gas turbine current, namely determining that the number of nodes of an input layer is 3.
(2) Determination of the number of output layer nodes:
the patent aims at predicting the disturbance load impedance value, so that the output layer is 1 node.
(3) Determining the number of hidden layer nodes and the network topology:
the number of nodes of the hidden layer is determined by the number of input nodes, the number of output nodes and an adjusting coefficient, namely, the network generalization capability is good, the regression analysis correlation coefficient is large, the training times are few, and the root mean square error is small, so that a satisfactory network structure is found; otherwise, adjusting the number of the neurons in the hidden layer, and repeatedly debugging until a satisfactory network structure is found. Since the initialization of the LSTM neural network is random, each network structure is simulated 30 times and averaged to reduce the error influence caused by the initialization. The training function of each layer of network adopts a self-adaptive learning rate momentum method, the activation function of each layer adopts a logarithmic sigmoid function (logsig), the output of the logarithmic sigmoid function is between 0 and 1, and convergence is facilitated.
The training results are shown in fig. 3 according to the above training rules. In the ten training processes, the number of hidden nodes is less in the initial stage, the fluctuation of the convergence step number is large, the hidden nodes gradually fall to an extreme value at 7 nodes, the training time is about 0.7 second at the moment, the number of the hidden nodes is changed, and the convergence step number is increased. Therefore, for the LSTM neural network, 7 nodes are set as the best choice for the hidden layer in this patent.
And judging whether the deviation value of the disturbance load impedance predicted value and the target value is within an allowable range or not according to the loss function value, if so, fixing the weight and the bias of each neuron in the neural network, determining an optimal network structure, and finishing the training. Otherwise, entering a reverse propagation process. And calculating the error between the predicted result and the true value of the impedance, reversely transmitting the error to a hidden layer of the neural network, and taking the calculated error of each layer as a basis for updating the weight and the bias by adopting a gradient descent method. Errors are continuously transmitted reversely in the whole training process, and weights and bias of each neuron are corrected, so that the neural network achieves a better impedance prediction effect, and the allowable deviation range of voltage and frequency in the technical specification of the microgrid energy management system is met.
And continuously and iteratively solving by a gradient descent method in the LSTM neural network training process to minimize the loss function, so that the target value and the deviation are continuously reduced. In this state, the optimal parameters of the model can be obtained, and the corresponding weight w and bias b are fixed as neuron parameters. In the LSTM neural network operation process, the process of searching the optimal solution by the gradient descent method is shown in FIG. 4. The red curve is an approximation process when the gradient descent method optimizes the disturbance load impedance value. When the minimum value point is solved, the weight w and the bias b corresponding to the point are the optimal solution of the neuron parameter. Fig. 5 shows that the LSTM neural network satisfies the requirement of 0.001 of the mean square error after 121 times of iterative training in the training process, and meanwhile, the convergence speed is increased due to the fact that the weight and the threshold are continuously adjusted in the initial training process, and a better disturbance load impedance value prediction level can be achieved.
If faults exist in the microgrid before the microgrid is put into operation, a microgrid fault information extraction model is set up by using the Petri network preferentially, and multi-mode prediction efficiency is improved. The basic elements of the Petri net are "place", "transition" and "Token", wherein place and transition are connected by a directed arc (arc). The library can describe the local state of the system through internally stored Token, and different states of the system are represented through dynamic change of the Token inside the library; the transition refers to a dynamic node caused by system change and describes an event for modifying the system state; the directed arcs are centralized embodiment of internal relations among elements of the system, and the evolution process of information dynamics in Petri can be clearly shown through direction.
As shown in fig. 6, Token is defined as the confidence level of the fault information in this example, the flow of Token represents the delivery of the fault, and different levels are represented by identifiers with different colors in the Petri chart; the failure mode of the system described by the library is represented by circles in a Petri graph, wherein ^ represents that the library contains Token and has a failure, and ^ represents that the library does not contain Token and has no failure, and ● represents that the library is in a failure result state; transitions are state variables describing the change in failure mode, represented by rectangles in Petri.
The Petri net defines twelve tuples as shown in formula (1).
S＝(P,T；I,O,K,C,M,Ω,α,f,H,U) (1)
Wherein P ═ P1,p2,...,pnThe node set of the limited non-empty storehouse (n > 0) can be used for representing the failure mode in the microgrid; t ═ T1,t2,...,tmThe (m is more than 0) is a finite non-empty transition node set, and t is a library node representing a fault state variable, and can reflect the replacement change of the system fault propagation process; i is a Petri network input matrix, and reflects the mapping from the library to the transition; the output matrix of the Petri network is represented by O, T multiplied by P, and the mapping of the transition to the library is reflected; k ═ K1,k2,...,knA limited set of Token-representing fault information; c is a color set, and different fault types are represented by a dye library; m ═ M1,m2,...,mn)TDistributing n-dimensional vectors, element m, for library identitiesi(i ═ 1, 2.., n) denotes the corresponding library piThe number of tokens and the color of the Token, the number of tokens indicates whether a fault occurs, the severity of the fault, and the number of paths causing the fault; omega-omega (omega)1,ω2,...,ωn)TRepresenting the influence of the input library place p on the transition rule t by an n-dimensional vector of the library place weight; α ═ α (α)1,α2,...,αn)TA fault event confidence n-dimensional vector; f ═ f1,f2,...,fn)TIs a set of library incident failure rates, where fiCorresponding to a library p in the fault information extraction modeliAnd represents piA set of failure rates of; h ═ λ1,λ2,...,λn)TIs a transition rule threshold m-dimensional vector; u ═ diag (μ)1,μ2,...,μm) For transition rule confidence matrix, element mujFor the transition rule tjThreshold value of (d), muj∈[0,1]。
The invention combines the LSTM neural network and the Petri network microgrid adaptive anti-islanding disturbance load impedance value prediction method, builds an impedance prediction model based on the LSTM neural network, and processes the disturbance load impedance value prediction problem on the time sequence; under the condition that a fault outgoing line exists, the fault information extraction efficiency and the micro-grid steady-state recovery level are improved through a micro-grid fault information extraction model based on the Petri grid, and the micro-grid fault information extraction efficiency and the micro-grid steady-state recovery level serve as data sources of the LSTM neural network under the fault condition.
In addition, the invention establishes impedance prediction models in different scenes in a targeted manner by analyzing the impedance characteristics in multiple scenes. A Long Short-Term Memory neural network (LSTM) is used as a time recursive neural network algorithm, is suitable for characteristics of a time sequence which need to be fully considered in the process of impedance value selection under the normal operation of a microgrid, if the microgrid is in a normal operation state before being put into use, the LSTM Memory mechanism is used for referring to an output value of a previous time sequence neuron, a gradient descent method is adopted for continuously updating a weight value and a bias in the process of reverse propagation, and the precision of predicting the dynamic impedance value of the input is improved. The Petri network provides an analysis method for parallel system research, is suitable for the change characteristic of discrete parallel impedance values under the condition of outlet line faults in the microgrid, and if fault outlet lines exist in the microgrid before the fault outlet lines are input, a microgrid fault information extraction model is built by utilizing the advantages of strong parallel asynchronism and capability of carrying out potential simulation of the Petri network and serves as a data source for predicting input disturbance load impedance values in the scene, dynamic impedance values are further predicted through the LSTM, and the built prediction model plays an active role in building a clean low-carbon energy supply system.
Finally, the invention can analyze the impedance characteristics under multiple scenes, can determine different prediction methods according to different running states of the microgrid, has better applicability to the prediction problem of disturbance load impedance value on a processing time sequence, responds to the change of supply and demand relations in real time, and can improve the fault information extraction efficiency and the steady-state recovery level of the microgrid. The self-adaptive impedance value prediction method plays an important role in improving the flexibility and reliability of the system and constructing a clean low-carbon energy supply system.
Claims (4)
1. A method for predicting a self-adaptive anti-islanding disturbance load impedance value of a micro-grid is characterized by comprising the following steps:
1) according to the state judgment of the anti-islanding device in the microgrid under different application scenes, whether the disturbance load impedance value is predicted or not is judged, and the application scenes of the anti-islanding device in the microgrid comprise:
(1) in the island operation maintenance scene, the anti-island device is in an input state, all converter switches in the island are disconnected after the input, and the island stops operating;
(2) under the condition of large power grid faults, a scene of partial island operation is allowed, the anti-island device state is removed, and planned island division is carried out;
(3) under the condition of large power grid fault and abnormal scenes occur in the island operation process, the anti-island device is in an input state, all converter switches in the island are disconnected after the input, and the island stops operating;
when the system is in the scenes of (1) and (3), forecasting the disturbance load impedance value;
2) when the micro-grid operates normally before the anti-islanding device is put into use, the disturbance load impedance value is predicted through the LSTM, and the prediction of the disturbance load impedance value through the LSTM specifically comprises the following steps:
21) constructing an anti-islanding disturbance load impedance prediction model based on an LSTM neural network, and determining the number of model layers and the number of neuron nodes in each layer;
22) training an anti-islanding disturbance load impedance prediction model, and determining optimal network parameters;
23) predicting a disturbance load impedance value according to a trained anti-islanding disturbance load impedance prediction model;
3) when the micro-grid fails before the anti-islanding device is put into use, a micro-grid fault information extraction model is established based on a Petri network, disturbance load impedance value prediction is carried out through an LSTM, Token is defined as the confidence level of fault information based on the Petri network, the flow of Token represents the transmission of faults, a state variable describing the change of a fault mode is used as a transition to establish the micro-grid fault information extraction model, and the micro-grid fault information extraction model is specifically expressed as follows:
S＝(P,T；I,O,K,C,M,Ω,α,f,H,U)
wherein, P ═ { P ═ P1,p2,...,pnN > 0 is a finite, non-empty pool node set representing a failure mode in the microgrid, T ═ T1,t2,...,tmIs a finite non-empty transition node set, tmThe method is characterized in that a library node representing a fault state variable is used for reflecting the replacement change of a system fault propagation process, I is a Petri net input matrix and reflects the mapping from a library to a transition, O is a Petri net output matrix and reflects the mapping from the transition to the library, and K is K and { K { (K) } T × P1,k2,...,kn-Token, representing a limited set of fault information, C a set of colors to represent different fault types by a color library, M ═ M (M)1,m2,...,mn)TDistributing n-dimensional vectors, element m, for library identitiesi(i ═ 1, 2.., n) denotes the corresponding library piThe number of tokens and the color of Token, the number of tokens indicates whether a fault occurs, the severity of the fault, and the number of paths causing the fault, and Ω ═ (ω ═ is1,ω2,...,ωn)TN-dimensional vector of library weight, representing the influence of the input library p on the transition rule t, α ═ α1,α2,...,αn)TFor the fault event confidence n-dimensional vector, f ═ f1,f2,...,fn)TIs a set of library incident failure rates, where fiCorresponding to a library p in the fault information extraction modeliAnd represents piSet of failure rates of (H ═ λ)1,λ2,...,λn)TFor the transition rule threshold m-dimensional vector, U ═ diag (μ)1,μ2,...,μm) For transition rule confidence matrix, element mujFor the transition rule tjA threshold value of (d), and μj∈[0,1]。
2. The method for predicting the self-adaptive anti-islanding disturbance load impedance value of the microgrid according to claim 1, wherein in the step 21), the anti-islanding disturbance load impedance prediction model takes photovoltaic current, fan current and combustion engine current data as input layers, the number of nodes of the input layers is 3, the disturbance load impedance value is taken as an output layer, and the number of nodes of the output layers is 1.
3. The method for predicting the self-adaptive anti-islanding disturbance load impedance value of the microgrid according to claim 1, characterized in that in the step 21), a training function of each hidden layer in the anti-islanding disturbance load impedance prediction model adopts a self-adaptive learning rate momentum method, an activation function of each layer adopts a logarithmic sigmoid function, and the number of nodes of the hidden layer is 7.
4. The method for predicting the self-adaptive anti-islanding disturbance load impedance value of the microgrid according to claim 1, characterized in that in the step 22), in a process of training an anti-islanding disturbance load impedance prediction model, a loss function is minimized through iteration of a gradient descent method, a target value and a deviation amount are reduced, and finally, an optimal network parameter of the model is obtained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911170555.XA CN110994673B (en) | 2019-11-26 | 2019-11-26 | Prediction method for micro-grid self-adaptive anti-islanding disturbance load impedance value |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911170555.XA CN110994673B (en) | 2019-11-26 | 2019-11-26 | Prediction method for micro-grid self-adaptive anti-islanding disturbance load impedance value |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110994673A CN110994673A (en) | 2020-04-10 |
| CN110994673B true CN110994673B (en) | 2021-07-20 |
Family
ID=70086926
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911170555.XA Active CN110994673B (en) | 2019-11-26 | 2019-11-26 | Prediction method for micro-grid self-adaptive anti-islanding disturbance load impedance value |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110994673B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111224427A (en) * | 2020-02-27 | 2020-06-02 | 上海电机学院 | Micro-grid self-adaptive anti-islanding disturbance load impedance value prediction method and device |
| CN112052233B (en) * | 2020-08-11 | 2021-03-30 | 杭州电子科技大学 | Multi-angle business process abnormity online detection method based on context awareness |
| CN112289385B (en) * | 2020-09-17 | 2022-08-09 | 西南交通大学 | Electrochemical impedance spectrum prediction method for high-power proton exchange membrane fuel cell stack |
| CN117590152B (en) * | 2023-11-21 | 2024-05-28 | 国网宁夏电力有限公司中卫供电公司 | Digital detection device for grid island of micro-grid integrated into power grid |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106849169A (en) * | 2017-03-20 | 2017-06-13 | 重庆大学 | A kind of simple micro-grid island detection method |
| CN106909989A (en) * | 2017-02-24 | 2017-06-30 | 国网河南省电力公司电力科学研究院 | A kind of grid disturbance Forecasting Methodology and device |
| CN106908132A (en) * | 2017-01-20 | 2017-06-30 | 山东科技大学 | A kind of method that strain gauge load cell failure is detected based on improved Petri net |
| CN107039976A (en) * | 2017-06-08 | 2017-08-11 | 山东鲁能智能技术有限公司 | It is a kind of to lead the intelligent anti-islanding system of the power distribution network passively combined and configuration operation method |
| CN109039739A (en) * | 2018-08-01 | 2018-12-18 | 南瑞集团有限公司 | A kind of power communication field failure diagnostic method improving fuzzy Petri |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10431980B2 (en) * | 2016-01-25 | 2019-10-01 | Enphase Energy, Inc. | Method and apparatus for increased energy harvest in a microgrid |
-
2019
- 2019-11-26 CN CN201911170555.XA patent/CN110994673B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106908132A (en) * | 2017-01-20 | 2017-06-30 | 山东科技大学 | A kind of method that strain gauge load cell failure is detected based on improved Petri net |
| CN106909989A (en) * | 2017-02-24 | 2017-06-30 | 国网河南省电力公司电力科学研究院 | A kind of grid disturbance Forecasting Methodology and device |
| CN106849169A (en) * | 2017-03-20 | 2017-06-13 | 重庆大学 | A kind of simple micro-grid island detection method |
| CN107039976A (en) * | 2017-06-08 | 2017-08-11 | 山东鲁能智能技术有限公司 | It is a kind of to lead the intelligent anti-islanding system of the power distribution network passively combined and configuration operation method |
| CN109039739A (en) * | 2018-08-01 | 2018-12-18 | 南瑞集团有限公司 | A kind of power communication field failure diagnostic method improving fuzzy Petri |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110994673A (en) | 2020-04-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110994673B (en) | Prediction method for micro-grid self-adaptive anti-islanding disturbance load impedance value | |
| Wang et al. | A review of applications of artificial intelligent algorithms in wind farms | |
| Hu et al. | Deep bidirectional recurrent neural networks ensemble for remaining useful life prediction of aircraft engine | |
| Su et al. | Deep belief network enabled surrogate modeling for fast preventive control of power system transient stability | |
| CN109242147A (en) | Signal fused fan condition prediction technique based on Bp neural network | |
| CN109657789A (en) | Gear case of blower failure trend prediction method based on wavelet neural network | |
| CN113141012B (en) | Power grid power flow regulation and control decision reasoning method | |
| CN117578466B (en) | Power system transient stability prevention control method based on dominant function decomposition | |
| Zhang et al. | Deep reinforcement learning for load shedding against short-term voltage instability in large power systems | |
| Zeng et al. | Distributed deep reinforcement learning-based approach for fast preventive control considering transient stability constraints | |
| CN111224427A (en) | Micro-grid self-adaptive anti-islanding disturbance load impedance value prediction method and device | |
| CN117039981A (en) | Large-scale power grid optimal scheduling method, device and storage medium for new energy | |
| Zeng et al. | A multiagent deep deterministic policy gradient-based distributed protection method for distribution network | |
| Li et al. | A multi-agent deep reinforcement learning-based “Octopus” cooperative load frequency control for an interconnected grid with various renewable units | |
| CN118174308A (en) | New energy power grid voltage active support control method based on multi-mode information | |
| CN114204546A (en) | Unit combination optimization method considering new energy consumption | |
| Prasad et al. | Online Identification of Self‐Organizing Fuzzy Neural Networks for Modeling Time‐Varying Complex Systems | |
| Yang et al. | Cloud-edge-end intelligence for fault-tolerant renewable energy accommodation in smart grid | |
| Xu et al. | Collaborative optimization of multi-energy multi-microgrid system: A hierarchical trust-region multi-agent reinforcement learning approach | |
| CN116362377A (en) | Large power grid region cooperative power flow regulation and control method based on multi-agent strategy gradient model | |
| Wei et al. | A lite cellular generalized neuron network for frequency prediction of synchronous generators in a multimachine power system | |
| Seo et al. | A study on modeling using big data and deep learning method for failure diagnosis of system | |
| Zhang et al. | Wind Speed Prediction with LSTM and BPNN Method | |
| Zhou et al. | Transient stability assessment of large-scale AC/DC hybrid power grid based on separation feature and deep belief networks | |
| Jawad et al. | Fault detection and location for VSC-HVDC systems with the coexistence of AC and DC Faults by using PSO-neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |